Six stages: build the library, plan, generate, fix, assemble, deliver. The library stage uses no tool at all and determines how well the rest goes. Planning is free in three of the four tools. Generation is where the tool choice actually happens, and the two handoffs that cause problems are both about framing — because three of Creative Studio's four modes inherit it and won't let you change it.
Most workflow advice for this is a list of tools. The useful version is a sequence, because the mistakes happen between stages rather than inside them.
Stage one — build the library
No tool, and it's the stage that determines everything downstream.
Every actor, location, style and recurring object gets created once and saved with a handle. You call the handle in prompts from then on rather than describing the thing again — which is what stops a face, a room or a look drifting between generations.
This happens before the script is locked, not after. A character you've written six episodes around is expensive to recast; a character created first is something you write toward.
The library stage uses no tool and decides how well every other stage goes.
Stage two — plan
Free, in three of the four tools, which makes it the cheapest stage to take seriously.
Plan First generates a storyboard and script preview without consuming credits, in Make a Custom Video, Talking Actors and URL → Ad. It's the only free step in any of those flows.
Use it for what it's actually good at: confirming the tool read your brief the way you meant it. On a prompt-first tool that fills in blanks, or on URL → Ad where there's no prompt field at all, that's the difference between catching a wrong interpretation now and paying for it.
Creative Studio has no equivalent, because there's nothing to interpret — you're specifying the shot rather than describing an outcome.
Stage three — generate
Where the tool choice happens, and it's a per-shot decision rather than a per-project one.
| What the shot is | Tool | Why |
|---|---|---|
| A face delivering a script | Talking Actors | Duration comes from the script, capped at sixty seconds. Emotion control places a beat at an exact word, which no prompt will match |
| A described scene you don't need to art-direct | Make a Custom Video | Five layers plus a scene prompt. Faster per shot, and unset layers get filled rather than skipped |
| A specific shot with blocking, coverage or camera movement | Creative Studio, Create mode | Direct model choice, up to 50 references, seven aspect ratios, and start and end frames |
| A product page becoming a vertical ad | URL → Ad | Three steps, no prompt field, 9:16 locked. Narrow on purpose and fast for exactly that |
| Movement transferred onto a character | Creative Studio, Motion mode | A driving video plus a character image. The prompt covers environment only |
Source: Hexcoded product documentation, checked September 2026. Tool capabilities change — verify against the product.
One balance covers all four, so switching mid-project costs a decision rather than a subscription. That's what makes per-shot choice viable at all.
The first handoff — framing
This is where productions lose time, and it happens between stage three and stage four.
Create mode offers seven aspect ratios. Edit, Extend and Motion each inherit it from the source clip and hide the selector entirely. So the framing of a chain is set by the first Create generation in it, and nothing downstream can change it.
Which means a shot you intend to edit or extend has its aspect ratio decided before you know whether you'll need to. And a project that needs both a vertical and a landscape cut needs two chains from two Create generations rather than one reframed at the end.
Where this falls short. That's irreversible within a chain rather than forever. You can always start a new Create generation. You just can't reframe the work you've already built on.
Framing is decided in the first generation of a chain. Three of four modes won't let you revisit it.
Stage four — fix
Two routes, and the second handoff is knowing which one applies.
Edit regenerates a clip against a description, conditioned on the source — so framing, light and identity carry forward rather than being re-guessed. It needs a source clip of three to ten seconds.
Extend appends footage to a clip you've already rendered, chaining across rounds while maintaining continuity.
The constraint worth knowing before stage three: a thirty-second single take cannot be edited as one piece, because it's outside Edit's source range entirely. So if a long take might need fixing, generating it in editable lengths is a stage-three decision rather than a stage-four discovery.
And three things can't be fixed this way at all — identity, lighting and performance. Each is a property of the whole take, so a described change alters more than you asked for. Those go back to regeneration.
Stage five — assemble
Outside the platform, and with one constraint that catches narrative producers.
Where a real human creator appears, only light edits are permitted — trimming, cropping, captions and subtitles, music, end-cards. Re-generating, morphing or re-voicing a face outside Hexcoded is prohibited, and it's listed as a prohibited use rather than merely a contract breach.
Colour, cuts, sound design and titles are all fine. Anything touching the performance has to happen on the platform.
For an advertising workflow that's rarely a problem. For a production with an external post house, it's a constraint to raise before the contract rather than at the grade.
Worth reviewing the assembled sequence before fixing individual shots, too. Editing a shot that gets cut is the most common wasted work in this whole sequence.
Stage six — deliver
Two things to know, and both are licence questions rather than technical ones.
You own the outputs, per the terms, and there's no watermark. There's also no AI label — any labelling required where you publish is your obligation as publisher, and every platform runs its own disclosure system.
And the scope is digital: paid ads and organic posts on online platforms, plus your own websites and apps. Television, print, outdoor and other offline use require a separate written agreement.
That last one belongs at stage six in this post and at stage one in your actual process. A campaign heading for broadcast needs that conversation before anything is generated.
The whole sequence
Six stages, with the two handoffs marked.
Build the library
Actors, locations, styles, props. Each created once and saved with a handle. No tool, and it determines everything downstream.
Plan, free
Plan First in Make a Custom Video, Talking Actors or URL → Ad. Confirms the tool read your brief the way you meant it, at no cost.
Generate — and set the framing here
Pick the tool per shot. Set aspect ratio and resolution now, because Edit, Extend and Motion inherit both. If you might edit a long take, generate it in editable lengths.
Fix — Edit for described changes, Extend for more footage
Edit needs a three-to-ten-second source. Identity, lighting and performance go back to regeneration instead.
Assemble — review the sequence before fixing shots
Light edits only where a real creator appears. And editing a shot you'll cut is the most common waste in the process.
Deliver, within the licensed scope
You own it, there's no watermark, and no AI label. Digital surfaces are covered; offline needs a separate agreement.
- Stage one uses no tool and decides how well the rest goes. Build the library first
- Planning is free in three of four tools. On anything that fills in blanks, use it every time
- Tool choice is per shot, not per project. One balance makes that viable
- Set the framing at stage three. Edit, Extend and Motion all inherit it and hide the selector
- Two aspect ratios means two chains, not one reframed at the end
- Edit needs a three-to-ten-second source. A thirty-second take can't be edited as one piece
- Identity, lighting and performance can't be fixed by editing. All three go back to regeneration
- Review the assembled sequence before fixing individual shots. Editing a shot you'll cut is pure waste
Six stages: build the library, plan, generate, fix, assemble, deliver. The library stage uses no tool and determines how well the rest goes, because saved references are what stop faces, rooms and looks drifting between generations.
A face delivering a script is Talking Actors. A described scene you don't need to art-direct is Make a Custom Video. A specific shot with blocking or camera movement is Creative Studio's Create mode. A product page becoming a vertical ad is URL → Ad.
At the generation stage, in Create mode. Edit, Extend and Motion all inherit it from the source clip and hide the selector, so framing is fixed by the first generation in a chain and can't be revisited downstream.
Only clips between three and ten seconds. A thirty-second single take is outside Edit's source range entirely, so if a long take might need fixing, generating it in editable lengths is a decision to make before you generate rather than after.
Where a real human creator appears, light edits only — trimming, cropping, captions, music, end-cards. Re-generating, morphing or re-voicing a face outside the platform is prohibited. Colour, cuts and sound design are fine; anything touching performance isn't.
Two licence questions. That the output is running on a digital surface, since television, print and outdoor need a separate written agreement. And that you've set the AI disclosure on whatever platform you're publishing to, since the platform adds no label and that obligation is yours.
Four tools, one sequence
One library and one credit balance across Creative Studio, Talking Actors, Make a Custom Video and URL → Ad. Pick the tool per shot, and every setting is visible before you generate.
Open Creative StudioMore on tools, features and how the platform fits together in Product.